Data-driven, Context-aware Human Fatigue Management in Traffic Control Centers

AutorLi, Fan
AbstraktTraffic control operators (TCOs) are at risk of human fatigue due to their heavy workload, their monotonous work, and the long duration of their shift work. Human fatigue can result in inefficiency, impaired alertness, and even a high probability of human error, and it has been identified as the primary cause of many traffic accidents. Thus, much research has focused on the management of human fatigue in transportation. However, strategies for managing human fatigue in transportation mainly focus on a single aspect of human fatigue, such as mental fatigue or physical fatigue. In addition, the existing strategies primarily center on prescriptive hours-of-work regulations. However, such regulations lack flexibility and fail to consider factors such as individual differences and the dynamic working conditions involved in traffic control. To fill this research gap, this study aims to develop context-aware human fatigue management strategies based on dynamic interactions between the contextual factors of human fatigue other than hours of work. In this study, data-driven approaches are proposed to prevent and mitigate human fatigue across traffic control centers (TCCs) and to develop context-aware fatigue management, which consists of four parts: task-driven causal factor analysis (TCFA), eye movement-based fatigue detection (EMFD), context-aware human fatigue prediction (CHFP) and data-driven intervention design (DID). Specifically, TCFA identifies and evaluates direct and indirect causal factors using a fatigue causal network that considers the multidimensional aspects of human fatigue. EMFD enables an objective and unobtrusive assessment of human fatigue by conducting gaze-bin analysis to obtain inputs for semisupervised bagged trees. EMFD paves the way for an alternative means of detecting human fatigue and enables the application of a low sampling rate eye tracker in TCCs. CHFP adopts a novel fatigue causal network and bagged tree techniques to predict human fatigue based on complex and interacting context data. Both dynamic work conditions and individual differences are considered in CHFP. To lower the occurrence of fatigue, a novel scheduling algorithm and a user requirement (UR)-driven framework for the redesign of traffic alarm systems are developed to achieve DID, which adaptively arranges work for operators by considering individual differences and work types and provides recommendations for reducing alarm fatigue. Several case studies were conducted in vessel traffic service centers to demonstrate and test the efficiency of the proposed context-aware human fatigue management approach. TCFA enabled both direct and indirect causal factor analysis. EMFD achieved an excellent accuracy of 89%, outperforming other classical methods, while CHFP achieved a high accuracy of 89%. Furthermore, DID, which adaptively arranges work for operators by considering individual differences and work types, showed that 27% of operators could be rearranged to reduce the possibility of human fatigue. In conclusion, this research develops a novel context-aware system for human fatigue management that advances the applications of artificial intelligence (AI) in human factor studies, provides insightful guidance for safety managers regarding the mitigation of human fatigue, and lays the foundation for further explorations of context-aware safety management.